Novel Land Cover Change Detection Method Based on k-Means Clustering and Adaptive Majority Voting Using Bitemporal Remote Sensing Images

Abstract

Land cover change detection (LCCD) based on bitemporal remote sensing images has become a popular topic in the field of remote sensing. Despite numerous methods promoted in recent decades, an improvement on the usability and performance of these methods has remained necessary. In this paper, a novel LCCD approach based on the integration of $k$ -means clustering and adaptive majority voting ( $k$ -means_AMV) techniques have been developed. The proposed $k$ -means_AMV method consists of three major techniques. First, to utilize the contextual information in an adaptive manner, an adaptive region around a central pixel is constructed by detecting the spectral similarity between the central pixel and its eight neighboring pixels. Second, when the extension for the adaptive region is terminated, the $k$ -means clustering method is applied to determine the label of each pixel within the adaptive region. Finally, an existing AMV technique is used to refine the label of the central pixel of the adaptive region. When change magnitude image (CMI) is scanned and processed in this manner, the label of each pixel in the CMI can be refined and the binary change detection map can be generated. Three image scenes related to different land cover change events are adapted to test the effectiveness and performance of the proposed $k$ -means_AMV approach. The results show that the proposed $k$ -means_AMV approach demonstrates better detection accuracies and visual performance than that of the several extensively used methods.